retro

CONTEXT LOADING

Before running a retro, load product.local.md for product context, quality standards, and PM process standards. If not configured, ask the user for product context and the feature being reviewed.

RETROSPECTIVE WORKFLOW

The Four Retro Questions

A useful product retro answers exactly four questions:

Q1: DID IT SOLVE THE PROBLEM? Was the PM right about the user need? Did the feature change what it was supposed to change?

Q2: DID WE BUILD IT AS INTENDED? Was the spec good enough? Did engineering deliver what was specified? Were there scope changes, quality issues, or delivery delays?

Q3: WERE THE METRICS RIGHT? Were the success metrics we defined measuring the right things? Did we have a failure threshold, and were we honest about it?

Q4: WHAT WOULD WE DO DIFFERENTLY? Not "what went wrong" -- but "what specific process change would we make if we started this feature again today?"

Data Required Before Running a Retro

OUTCOME DATA (4-12 weeks post-launch):

• Adoption: % of target users who have used the feature at least once
• Engagement: how often? for how long? (if applicable)
• Support impact: did support tickets about this area increase or decrease?
• NPS impact: has there been movement in relevant NPS questions?
• Business metric: did the primary success metric move?

DELIVERY DATA:

• On time or late? If late: by how long?
• Were there scope changes during development? What was added or cut?
• Were there significant spec ambiguity issues flagged by engineering?

TEAM DATA (optional but valuable):

• Did engineering flag any spec quality issues?
• Did design flag any requirement gaps during design review?
• Did CS/Support flag anything unexpected post-launch?

Retro Output Format

PRODUCT RETROSPECTIVE: [Feature name]
Shipped: [Date] | Retro date: [Date] | PM: [Name]
================================================================

Q1: DID IT SOLVE THE PROBLEM?

Original problem:   [From the spec -- what were we trying to solve?]
Target outcome:     [From the spec -- what did we say success looked like?]

Evidence:
[PASS/WARN/FAIL] [Metric]: [actual result] vs. [target]
[PASS/WARN/FAIL] [Metric]: [actual result] vs. [target]

Verdict: [SOLVED / PARTIALLY SOLVED / NOT SOLVED / TOO EARLY TO TELL]
Summary: [2-3 sentences: what the data shows about whether the
          feature delivered on its intent]

Q2: DID WE BUILD IT AS INTENDED?

[PASS/WARN/FAIL] Delivered on time: [On time / [N] weeks late -- reason]
[PASS/WARN/FAIL] Spec quality:     [No major ambiguity / [Issue reported]]
[PASS/WARN/FAIL] Scope maintained: [No changes / Changes: describe]
[PASS/WARN/FAIL] Quality (bugs):   [Bug rate in first 30 days]
[PASS/WARN/FAIL] ACs met:          [All ACs verified / Outstanding: list]

Q3: WERE THE METRICS RIGHT?

For each success metric defined in the spec/PRD:
- Was it measurable? (Could we actually get the data?)
- Was it sensitive? (Did it move in response to the feature?)
- Was it the right thing to measure? (Did it reflect user value?)
- Did we have a failure threshold? Did we use it?

Metric quality rating: [HIGH / MEDIUM / LOW]
Learning: [What we should measure differently next time]

Q4: WHAT WOULD WE DO DIFFERENTLY?

Process improvement [N]:
  What happened:  [Specific -- not "better communication"]
  What to change: [Specific rule or step to add/change]
  How to apply:   [On the next feature: [specific action]]

FOLLOW-ON ACTIONS (from retro findings)
[N]. [Action] -- [Owner] -- [Due date]

PRODUCT.LOCAL.MD UPDATES RECOMMENDED
[Any quality rules to add to the configuration based on this retro]
================================================================

Process Improvement Format

Rule: Every "what went wrong" must be converted into a specific, testable process change -- not a vague intention.

WEAK: "Write better specs" STRONG: "Every AC must be independently testable. If an AC contains 'and', split it before REVIEW status."

WEAK: "Communicate better with engineering" STRONG: "Engineering lead must confirm architecture notes in PRD before the PRD moves to REVIEW status."

WEAK: "Define success metrics earlier" STRONG: "Primary metric and failure threshold must be in the spec before it enters sprint planning. No spec without metrics."

COMPLEMENTARY WORKFLOWS

This skill handles product retrospectives. For related PM workflows:

• Reviewing metrics that inform the retro: use official /metrics-review
• Communicating retro findings to stakeholders: use official /stakeholder-update
• Updating the roadmap based on retro learnings: use official /roadmap-update

NEVER DO THESE

• NEVER run a retro without outcome data -- opinions without data produce blame, not learning
• NEVER produce a retro that only documents what went wrong -- what went well is equally important (protect what works)
• NEVER produce a vague process improvement ("communicate better") -- convert every improvement into a specific, enforceable rule
• NEVER skip the metrics quality section -- bad metrics are as dangerous as no metrics; they create false confidence
• NEVER close a retro without at least one product.local.md update -- if nothing changes, the retro was a waste of time

ALL OUTPUTS REQUIRE REVIEW BY THE PM AND ENGINEERING LEAD.